CN101599077B - Method for retrieving three-dimensional object - Google Patents
Method for retrieving three-dimensional object Download PDFInfo
- Publication number
- CN101599077B CN101599077B CN2009100884162A CN200910088416A CN101599077B CN 101599077 B CN101599077 B CN 101599077B CN 2009100884162 A CN2009100884162 A CN 2009100884162A CN 200910088416 A CN200910088416 A CN 200910088416A CN 101599077 B CN101599077 B CN 101599077B
- Authority
- CN
- China
- Prior art keywords
- view
- view collection
- collection
- search object
- dimensional
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Abstract
The invention discloses a method for retrieving a three-dimensional object, comprising the following steps: carrying out bipartite graph maximum matching on the three-dimensional retrieval object view sets and the view sets in a database, filtering the view sets not satisfying the preset matching conditions in the database to obtain the remaining view sets; carrying out bipartite graph optimal matching on the three-dimensional retrieval object view sets and each remaining view sets to acquire the distances between the three-dimensional retrieval object view sets and each remained view sets; sequencing each remaining view sets and outputting each sequenced remaining view sets as the retrieval results. The embodiment of the invention adopts pretreatment of statistical clustering, combines bipartite graph maximum matching and bipartite graph optimal matching to acquire the distances between the three-dimensional retrieval object view sets and each remaining view sets, thus improving the correctness of three-dimensional object retrieval, and simultaneously causing the three-dimensional object retrieval to be capable of not depending on collecting environment information, and the three-dimensional object retrieval method to be more vastly applied.
Description
Technical field
The present invention relates to field of image search, particularly a kind of method of retrieving three-dimensional objects.
Background technology
Three dimensional object is becoming important computer media data type because it is true, three-dimensional, and its quantity increases rapidly on internet and PC.Therefore the user has the demand that the three dimensional object data message is retrieved, and promptly obtains required information from the three dimensional object data of magnanimity.
The retrieving three-dimensional objects method has statistical information based on model, based on geometry with based on these three kinds of modes of model view.These methods all are to depend on existing model information, by the geometric model that builds, calculate the information that needs, even if based on the retrieval mode of view, also be according to model information, and then obtain the view of all angles.Because the complexity of model is directly relevant with the complexity of actual object or scene, the model that places one's entire reliance upon can cause the increase of computation complexity, and for complex scene, modeling is very difficult even can't realize.
Represent at present and the technology of playing up has all had important development based on the three dimensional object of image, particularly intensive many views Sampling techniques, be a kind of fast development and the three-dimensional description technique that has much prospect, many views collection of three dimensional object is generally with the following method: place a video camera array to take three dimensional object, obtain describing the two dimensional image set of this object, also become a very promising research direction based on the retrieving three-dimensional objects of many views.
In the method for existing retrieving three-dimensional objects, mostly need to obtain to gather the information of environment, as information such as video camera array distribution modes, and to gather environment be fixed single mostly, as the video camera array of annular spread uniformly-spaced etc.In the retrieving of three dimensional object, gather obtaining of environmental information owing to rely on, make the application of retrieving three-dimensional objects method have certain limitation.
Summary of the invention
The embodiment of the invention provides a kind of method of retrieving three-dimensional objects, and described method comprises:
Obtain the view collection of the three-dimensional search object of user's input;
View collection in described three-dimensional search object view collection and the database is carried out the bipartite graph maximum match, do not satisfy the view collection of preset matching condition in the filtering database, obtain remaining the view collection;
View collection to described three-dimensional search object is added up cluster, obtains the statistics cluster result of described three-dimensional search object view collection;
Each described residue view collection is added up cluster, obtain the statistics cluster result of each described residue view collection;
According to the statistics cluster result of described three-dimensional search object view collection and the statistics cluster result of each described residue view collection, described three-dimensional search object view collection and each described residue view collection are carried out the bipartite graph Optimum Matching, obtain the distance between described three-dimensional search object view collection and each the described residue view collection;
Distance according between described three-dimensional search object view collection and each the described residue view collection sorts to each described residue view collection, and each the described residue view collection after the ordering is exported as result for retrieval.
The embodiment of the invention is in the process of retrieving three-dimensional objects, by adopting the pre-service of statistics cluster, in conjunction with bipartite graph maximum match and Optimum Matching, obtain the distance between the view collection in three-dimensional search object view collection and the database, improved the accuracy of retrieving three-dimensional objects, make the retrieval of three dimensional object can not rely on the information of gathering environment simultaneously, the method for retrieving three-dimensional objects can be applied even more extensively.
Description of drawings
Fig. 1 is the method flow diagram of the retrieving three-dimensional objects that provides in the embodiment of the invention 1.
Embodiment
For making the purpose, technical solutions and advantages of the present invention clearer, embodiment of the present invention is described further in detail below in conjunction with accompanying drawing.
Embodiment 1
The embodiment of the invention provides a kind of method of retrieving three-dimensional objects, and referring to Fig. 1, this method comprises:
101: the view collection that obtains the three-dimensional search object of user's input.
Concrete, the view collection of the three dimensional object of the needs retrieval of user's input is one group of two dimension view of this three dimensional object of expression.
102: the view collection to this three-dimensional search object is added up cluster, obtains the statistics cluster result of three-dimensional search object view collection.
Concrete, when the view collection is added up cluster, need earlier the view collection to be carried out Feature Extraction, what extract usually is the bottom visual signature of view, utilize nothing supervision or semi-supervised learning method that this view set is divided into the several views subclass then, each view subclass the inside comprises one group of visually similar image.As a view collection X={x
1, x
2..., x
n, x wherein
iFor the width of cloth two dimension view among the X, when carrying out cluster, be about to and x
iThe image x of visually similar image
p, x
q..., x
tBe divided into a view subset X
1={ x
i, x
p, x
q..., x
t, view collection X={x thus
1, x
2..., x
nAdd up cluster after, obtain k view subclass, i.e. X={X
1, X
2..., X
K, and the view in each view subclass all is visually similar view.Preferably, the value of k is between the 3-6.
103: the view collection in this three-dimensional search object view collection and the database is carried out the bipartite graph maximum match, do not satisfy the view collection of preset matching condition in the filtering database, obtain remaining the view collection.
Concrete, the method for maximum match of carrying out bipartite graph is as follows: establish X={x
1, x
2..., x
nThe time three-dimensional search object the view collection, Y
k={ y
1, y
2..., y
mIt is the view set of a three dimensional object in the database.Successively to X and Y
kIn view use the bipartite graph maximum match, do not satisfy the view collection of preset matching condition in the filtering database, obtain satisfying in the database each pre-conditioned residue view collection, wherein, this is pre-conditioned to be: with the similarity of this three-dimensional search object view collection greater than preset threshold value.
Preferably, can use the Hungarian algorithm.If the view x among the X
iAnd Y
kIn view y
jSimilarity greater than threshold value T, think that then this two width of cloth figure is relevant, connects this two width of cloth view with a sideline.Wherein, the value of threshold value T can be adjusted according to the number of match views.Under the constraint that view mates one to one, can obtain maximum match subgraph M, if when total number l in sideline is greater than default matching threshold (as n/2) among the maximum match subgraph M, then think Y
kSimilar with X.Otherwise total number l in sideline is less than default matching threshold among the maximum match subgraph M, then thinks Y
kWith the X dissmilarity, filtering Y
k, continue other view collection in the comparison database.
By said method as can be known, the computation complexity of the maximum match of bipartite graph is O ((n+m) * l), and wherein l is the number in the sideline among the maximum match subgraph M.By step 103, with regard to filtering with the incoherent view set of the three-dimensional search object of user input, the residue view collection that obtains is also just relevant with this three-dimensional search object.
104: each above-mentioned residue view collection is added up cluster, obtain the statistics cluster result of each residue view collection.
It is identical with method in 102 that each residue view collection in the database is added up the method for cluster, repeat no more, view collection in the database is added up cluster can carry out in advance, can directly extract the statistics cluster result of each residue view collection thus when retrieval from database.
105: according to the statistics cluster result of this three-dimensional search object view collection and the statistics cluster result of each residue view collection, this three-dimensional search object view collection and this each residue view collection are carried out the bipartite graph Optimum Matching, obtain the distance between this three-dimensional search object view collection and each this residue view collection.
Concrete, the method for Optimum Matching of carrying out bipartite graph is as follows:
1) obtain the cluster centre of view subclass, with the proper vector mean value of the view that belongs to this subclass as this subclass center, for X that is divided into K subclass and Y
k, obtain its subclass center and be designated as { x
1, x
2..., x
K, { y
1, y
2..., y
K, the computation complexity of step 1) is O ((n+m) * K).
2) to cluster centre X
c={ x
C1, x
C2..., x
CKAnd Y
Kc={ y
C1, y
C2..., y
CKTwo set, constitute bipartite graph G
c={ X
c, Y
Kc, E
c, use the Kuhn-Munkres algorithm, under the constraint of mating one to one, can obtain Optimum Matching subgraph M
cThereby, obtain X
c={ x
C1, x
C2..., x
CKAnd Y
Kc={ y
C1, y
C2..., y
CKThe corresponding relation of each view of lining, just view set X and Y
kIn the corresponding relation of each subclass, with Y
kAccording to resequencing, obtain X={X with the corresponding relation of X
1, X
2..., X
K, Y
k={ Y
k 1, Y
k 2..., Y
k K, X wherein
iAnd y
k iCorresponding.Step 2) computation complexity is O (K
4).
3) according to each to corresponding subset X
iAnd Y
k iConstitute a bipartite graph G
i={ X
i, Y
k i, E
i, the algorithm of use Optimum Matching as the Kuhn-Munkres algorithm, carries out the setting of weights to each limit, and concrete, in embodiments of the present invention, the Euclidean distance of two width of cloth view feature vectors is set to the weights on limit.Under the constraint of coupling one to one, try to achieve the subgraph of weights minimum, as the Optimum Matching of this bipartite graph, and to weights summation obtains subset X
iAnd Y
k iDistance:
Wherein,
n
i, m
iBe respectively X
iAnd Y
k iThe view number.
Further, corresponding subclass distance is sued for peace, obtain view set X and Y
kDistance:
In the step of the distance of two view collection of aforementioned calculation, as can be known since in intensive sampling the number n of view and m much larger than k, the statistics cluster is relative very little with computation complexity in the views registered process, can ignore, reduced the complexity of Optimum Matching thus, therefore, the present invention adopts the cluster pre-service to improve the efficient of system.
106: be somebody's turn to do the distance that remains between the view collection according to this three-dimensional search object view collection and each, each this residue view collection is sorted, each this residue view collection after the ordering is exported as result for retrieval.
Have foregoing description as can be known, the present invention adopts earlier the bipartite graph maximum match by the statistics cluster, the view collection not high with the searching object degree of correlation in the filtering database, and then adopt the bipartite graph Optimum Matching, further retrieval and coupling of view collection from database.The embodiment of the invention is by adopting the maximum match and the Optimum Matching of statistics cluster and bipartite graph, and feasible retrieving three-dimensional objects based on many views collection can not rely on the information of gathering environment, and the accuracy of retrieving three-dimensional objects also is improved simultaneously.
The embodiment of the invention is in the process of retrieving three-dimensional objects, by adopting the pre-service of statistics cluster, in conjunction with bipartite graph maximum match and Optimum Matching, obtain the distance between the view collection in three-dimensional search object view collection and the database, improved the accuracy of retrieving three-dimensional objects, make the retrieval of three dimensional object can not rely on the information of gathering environment simultaneously, the method for retrieving three-dimensional objects can be applied even more extensively.
The embodiment of the invention can utilize software to realize that corresponding software programs can be stored in the storage medium that can read, for example, and in the hard disk of router, buffer memory or the CD.
The above only is preferred embodiment of the present invention, and is in order to restriction the present invention, within the spirit and principles in the present invention not all, any modification of being done, is equal to replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (4)
1. the method for a retrieving three-dimensional objects is characterized in that, described method comprises:
Obtain the view collection of the three-dimensional search object of user's input;
View collection in described three-dimensional search object view collection and the database is carried out the bipartite graph maximum match, do not satisfy the view collection of preset matching condition in the filtering database, obtain remaining the view collection;
View collection to described three-dimensional search object is added up cluster, obtains the statistics cluster result of described three-dimensional search object view collection;
Each described residue view collection is added up cluster, obtain the statistics cluster result of each described residue view collection;
According to the statistics cluster result of described three-dimensional search object view collection and the statistics cluster result of each described residue view collection, described three-dimensional search object view collection and each described residue view collection are carried out the bipartite graph Optimum Matching, obtain the distance between described three-dimensional search object view collection and each the described residue view collection;
Distance according between described three-dimensional search object view collection and each the described residue view collection sorts to each described residue view collection, and each the described residue view collection after the ordering is exported as result for retrieval.
2. the method for retrieving three-dimensional objects according to claim 1 is characterized in that, described described three-dimensional search object view collection and each described residue view collection is carried out also comprising before the bipartite graph Optimum Matching:
The view collection of described three-dimensional search object is carried out the extraction of bottom visual signature; And the view collection of described three-dimensional search object is added up cluster according to the feature of extracting, obtain the statistics cluster result of described three-dimensional search object view collection;
Each described residue view collection is carried out the extraction of bottom visual signature; And each described residue view collection is added up cluster according to the feature of extracting, obtain the statistics cluster result of each described residue view collection.
3. the method for retrieving three-dimensional objects according to claim 1 is characterized in that, described preset matching condition is: with the similarity of described three-dimensional search object view collection greater than preset threshold value.
4. the method for retrieving three-dimensional objects according to claim 1, it is characterized in that, described described three-dimensional search object view collection and each described residue view collection are carried out the bipartite graph Optimum Matching, obtain the distance between described three-dimensional search object view collection and each the described residue view collection, comprising:
Obtain first cluster centre of the view subclass of described three-dimensional search object, and obtain second cluster centre of described residue view subclass;
Calculate the corresponding relation of described first cluster centre and described second cluster centre,, carry out corresponding with described residue view subclass the view subclass of described three-dimensional search object according to described corresponding relation;
According to the view subclass and the described residue view subclass of the described three-dimensional search object of mutual correspondence, calculate the distance between described three-dimensional search object view collection and the described residue view collection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009100884162A CN101599077B (en) | 2009-06-29 | 2009-06-29 | Method for retrieving three-dimensional object |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009100884162A CN101599077B (en) | 2009-06-29 | 2009-06-29 | Method for retrieving three-dimensional object |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101599077A CN101599077A (en) | 2009-12-09 |
| CN101599077B true CN101599077B (en) | 2011-01-05 |
Family
ID=41420525
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2009100884162A Expired - Fee Related CN101599077B (en) | 2009-06-29 | 2009-06-29 | Method for retrieving three-dimensional object |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101599077B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101894267B (en) * | 2010-07-06 | 2012-07-18 | 清华大学 | Three-dimensional object characteristic view selection method |
| CN102073738B (en) * | 2011-01-20 | 2013-04-17 | 清华大学 | Intelligent retrieval view selection-based three-dimensional object retrieval method and device |
| CN104298758A (en) * | 2014-10-22 | 2015-01-21 | 天津大学 | Multi-perspective target retrieval method |
| CN105243139B (en) * | 2015-10-10 | 2018-10-23 | 天津大学 | A kind of method for searching three-dimension model and its retrieval device based on deep learning |
| CN106503270B (en) * | 2016-12-09 | 2020-02-14 | 厦门大学 | 3D target retrieval method based on multi-view and bipartite graph matching |
| CN108717424B (en) * | 2018-04-25 | 2021-06-11 | 鹰霆（天津）科技有限公司 | Three-dimensional model retrieval method based on decomposition type graph matching |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040249809A1 (en) * | 2003-01-25 | 2004-12-09 | Purdue Research Foundation | Methods, systems, and data structures for performing searches on three dimensional objects |
| CN101281545A (en) * | 2008-05-30 | 2008-10-08 | 清华大学 | Three-dimensional model search method based on multiple characteristic related feedback |
| CN101458714A (en) * | 2008-12-30 | 2009-06-17 | 清华大学 | Three-dimensional model search method based on precision geodesic |
-
2009
- 2009-06-29 CN CN2009100884162A patent/CN101599077B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040249809A1 (en) * | 2003-01-25 | 2004-12-09 | Purdue Research Foundation | Methods, systems, and data structures for performing searches on three dimensional objects |
| CN101281545A (en) * | 2008-05-30 | 2008-10-08 | 清华大学 | Three-dimensional model search method based on multiple characteristic related feedback |
| CN101458714A (en) * | 2008-12-30 | 2009-06-17 | 清华大学 | Three-dimensional model search method based on precision geodesic |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101599077A (en) | 2009-12-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gu et al. | A review on 2D instance segmentation based on deep neural networks | |
| Zheng et al. | SIFT meets CNN: A decade survey of instance retrieval | |
| Meng et al. | Video summarization via multi-view representative selection | |
| ElAlami | A novel image retrieval model based on the most relevant features | |
| Chum | Large-scale discovery of spatially related images | |
| CN101599077B (en) | Method for retrieving three-dimensional object | |
| CN101853486B (en) | Image copying detection method based on local digital fingerprint | |
| JP4937395B2 (en) | Feature vector generation apparatus, feature vector generation method and program | |
| CN105205135A (en) | 3D (three-dimensional) model retrieving method based on topic model and retrieving device thereof | |
| Papadopoulos et al. | Image clustering through community detection on hybrid image similarity graphs | |
| Petkos et al. | Graph-based multimodal clustering for social event detection in large collections of images | |
| Su et al. | Scalable deep learning logo detection | |
| Suwanwimolkul et al. | Learning of low-level feature keypoints for accurate and robust detection | |
| Zeng et al. | Amorphous region context modeling for scene recognition | |
| Gao et al. | Event video mashup: From hundreds of videos to minutes of skeleton | |
| Wang et al. | Action recognition by latent duration model | |
| Hu et al. | STRNN: End-to-end deep learning framework for video partial copy detection | |
| Priya et al. | Optimized content based image retrieval system based on multiple feature fusion algorithm | |
| Deng et al. | Visualisation and comparison of image collections based on self-organised maps | |
| Vandaele et al. | Mining topological structure in graphs through forest representations | |
| Sun et al. | A novel region-based approach to visual concept modeling using web images | |
| Bouachir et al. | Fuzzy indexing for bag of features scene categorization | |
| Yang et al. | Saliency regions for 3D mesh abstraction | |
| Ahmadi et al. | Tags Re-ranking Using Multi-level Features in Automatic Image Annotation | |
| Liu et al. | VIEW GRAPH CONSTRUCTION FOR LARGE-SCALE UAV IMAGES: AN EVALUATION OF STATE-OF-THE-ART METHODS |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20110105 Termination date: 20140629 |
|
| EXPY | Termination of patent right or utility model |